📌 TOPINDIATOURS Hot ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-Imp

The concept of AI self-improvement has been a hot topic in recent research circles, with a flurry of papers emerging and prominent figures like OpenAI CEO Sam Altman weighing in on the future of self-evolving intelligent systems. Now, a new paper from MIT, titled “Self-Adapting Language Models,” introduces SEAL (Self-Adapting LLMs), a novel framework that allows large language models (LLMs) to update their own weights. This development is seen as another significant step towards the realization of truly self-evolving AI.

The research paper, published yesterday, has already ignited considerable discussion, including on Hacker News. SEAL proposes a method where an LLM can generate its own training data through “self-editing” and subsequently update its weights based on new inputs. Crucially, this self-editing process is learned via reinforcement learning, with the reward mechanism tied to the updated model’s downstream performance.

The timing of this paper is particularly notable given the recent surge in interest surrounding AI self-evolution. Earlier this month, several other research efforts garnered attention, including Sakana AI and the University of British Columbia’s “Darwin-Gödel Machine (DGM),” CMU’s “Self-Rewarding Training (SRT),” Shanghai Jiao Tong University’s “MM-UPT” framework for continuous self-improvement in multimodal large models, and the “UI-Genie” self-improvement framework from The Chinese University of Hong Kong in collaboration with vivo.

Adding to the buzz, OpenAI CEO Sam Altman recently shared his vision of a future with self-improving AI and robots in his blog post, “The Gentle Singularity.” He posited that while the initial millions of humanoid robots would need traditional manufacturing, they would then be able to “operate the entire supply chain to build more robots, which can in turn build more chip fabrication facilities, data centers, and so on.” This was quickly followed by a tweet from @VraserX, claiming an OpenAI insider revealed the company was already running recursively self-improving AI internally, a claim that sparked widespread debate about its veracity.

Regardless of the specifics of internal OpenAI developments, the MIT paper on SEAL provides concrete evidence of AI’s progression towards self-evolution.

Understanding SEAL: Self-Adapting Language Models

The core idea behind SEAL is to enable language models to improve themselves when encountering new data by generating their own synthetic data and optimizing their parameters through self-editing. The model’s training objective is to directly generate these self-edits (SEs) using data provided within the model’s context.

The generation of these self-edits is learned through reinforcement learning. The model is rewarded when the generated self-edits, once applied, lead to improved performance on the target task. Therefore, SEAL can be conceptualized as an algorithm with two nested loops: an outer reinforcement learning (RL) loop that optimizes the generation of self-edits, and an inner update loop that uses the generated self-edits to update the model via gradient descent.

This method can be viewed as an instance of meta-learning, where the focus is on how to generate effective self-edits in a meta-learning fashion.

A General Framework

SEAL operates on a single task instance (C,τ), where C is context information relevant to the task, and τ defines the downstream evaluation for assessing the model’s adaptation. For example, in a knowledge integration task, C might be a passage to be integrated into the model’s internal knowledge, and τ a set of questions about that passage.

Given C, the model generates a self-edit SE, which then updates its parameters through supervised fine-tuning: θ′←SFT(θ,SE). Reinforcement learning is used to optimize this self-edit generation: the model performs an action (generates SE), receives a reward r based on LMθ′’s performance on τ, and updates its policy to maximize the expected reward.

The researchers found that traditional online policy methods like GRPO and PPO led to unstable training. They ultimately opted for ReST^EM, a simpler, filtering-based behavioral cloning approach from a DeepMind paper. This method can be viewed as an Expectation-Maximization (EM) process, where the E-step samples candidate outputs from the current model policy, and the M-step reinforces only those samples that yield a positive reward through supervised fine-tuning.

The paper also notes that while the current implementation uses a single model to generate and learn from self-edits, these roles could be separated in a “teacher-student” setup.

Instantiating SEAL in Specific Domains

The MIT team instantiated SEAL in two specific domains: knowledge integration and few-shot learning.

• Knowledge Integration: The goal here is to effectively integrate information from articles into the model’s weights.
• Few-Shot Learning: This involves the model adapting to new tasks with very few examples.

Experimental Results

The experimental results for both few-shot learning and knowledge integration demonstrate the effectiveness of the SEAL framework.

In few-shot learning, using a Llama-3.2-1B-Instruct model, SEAL significantly improved adaptation success rates, achieving 72.5% compared to 20% for models using basic self-edits without RL training, and 0% without adaptation. While still below “Oracle TTT” (an idealized baseline), this indicates substantial progress.

For knowledge integration, using a larger Qwen2.5-7B model to integrate new facts from SQuAD articles, SEAL consistently outperformed baseline methods. Training with synthetically generated data from the base Qwen-2.5-7B model already showed notable improvements, and subsequent reinforcement learning further boosted performance. The accuracy also showed rapid improvement over external RL iterations, often surpassing setups using GPT-4.1 generated data within just two iterations.

Qualitative examples from the paper illustrate how reinforcement learning leads to the generation of more detailed self-edits, resulting in improved performance.

While promising, the researchers also acknowledge some limitations of the SEAL framework, including aspects related to catastrophic forgetting, computational overhead, and context-dependent evaluation. These are discussed in detail in the original paper.

Original Paper: https://arxiv.org/pdf/2506.10943

Project Site: https://jyopari.github.io/posts/seal

Github Repo: https://github.com/Continual-Intelligence/SEAL

The post MIT Researchers Unveil “SEAL”: A New Step Towards Self-Improving AI first appeared on Synced.

🔗 Sumber: syncedreview.com

📌 TOPINDIATOURS Breaking ai: Barrage of Emails From AI Politics Platform Defeats C

Late in 2023, the corporate world was abuzz with the utopian promise of AI — especially about the tech’s effects on the environment.

From today’s vantage point, some of the corporate claims floating around back then sound patently absurd: Google insisted that AI had the potential to “mitigate 5-10 percent of global greenhouse gas emissions,” while Microsoft enthused that the new tech would “accelerate the discovery and development of sustainability solutions.” Even academics played along. Jim Bellingham of Johns Hopkins’ Institute for Assured Autonomy claimed that “AI-powered robots” and satellites would form part of a system helping to “reduce the carbon that is released into the atmosphere.”

So far, reality has demonstrated the exact opposite. Not only are electrical demands of data centers supercharging our carbon emissions, but the technology itself is now being used to actively resist climate regulations.

According to new reporting by the Los Angeles Times, an eco-friendly initiative to phase out the use of gas-powered appliances in Southern California was defeated thanks to a campaign weaponizing a suite of AI software. The regulation, proposed by the South Coast Air Quality Management District (AQMD), would have gradually phased out household water heaters and furnaces powered with natural gas, as part of an effort to limit the emissions of smog-causing nitrogen oxide.

As the initiative proceeded through AQMD’s board in summer of 2025, the agency became inundated with tens of thousands of seemingly organic emails voicing opposition to the regulations. As the LA Times notes, the flood of angry public comments got a significant boost from CiviClick, an AI platform which bills itself as a “Disruptive Digital Advocacy Software.”

According to records requests, more than 20,000 public comments directed at AQMD’s initiative were drummed up via CiviClick.

While it isn’t clear where AI was deployed in the campaign, CiviClick boasts a number of solutions bundled as “Grassroots Advocacy Software.” These include auto-filled webforms, as well as video and photo messages to elected officials. Most crucially CiviClick advertises an “AI powered automatic message generator,” which comes with “auto-randomizing” messaging, “unlimited subject lines and message bodies,” and “smart targeting ability.”

In the face of such vast public opposition — or what appeared to be, at the time — the AQMD board voted 7-5 to reject the measure, which would have slowly raised the cost of gas-powered appliances through small surcharges.

The email campaign’s mastermind, Matt Klink, went so far as to say CiviClick made “the ultimate difference” in shooting down the environmental safeguards.

“We did two separate rounds of outreach, and they met the targets in both rounds early,” Klink told political consulting publication Campaigns and Elections in a sponsored interview. “AQMD staff are not used to getting tens of thousands of emails so it [CiviClick] made a massive difference in turning the tide.”

Though Klink claimed that CiviClick was simply an email tool used to connect voters to AQMD staff en-masse, an investigation showed otherwise. As the LA Times notes, at least three people contacted afterward said they weren’t aware that CiviClick had sent comments to AQMD on their behalf.

As sources within AQMD told the newspaper, the onslaught definitely had an impact on the board’s decision, which in-turn played into the hands of the gas industry. (Which, for its part, had been waging a legal battle against the environmental agency since December of 2024.)

As one staffer told the LA Times, the usual amount of public comments per agenda item can be counted on one hand.

Notably, Klink declined to tell the LA Times who had funded the CiviClick campaign. His public affairs company, Klink Campaigns, is a partner at California Strategies, which the paper notes is one of the largest lobbying firms in the state. Its clients include corporate landlord groups, energy conglomerates, and Fortune 500 energy company Sempra, owner of the massive Southern California Gas Company.

While astroturfing is nothing new in US politics, AI-powered astroturfing marks an ominous escalation. On top of signaling a horrifying turn for civic engagement, it opens up a tremendous can of worms for political campaigning across the US. As long as AI can fabricate thousands of constituents on demand, corporate interests no longer need the pretense of democracy — they can just rent it.

Do you know anything about the use of AI in politics? Send us a tip: tips@futurism.com.

More on AI: AI Is Incredibly Good at Changing Voters’ Minds, New Research Finds — With an Incredible Caveat

The post Barrage of Emails From AI Politics Platform Defeats Clean Air Initiative appeared first on Futurism.

🔗 Sumber: futurism.com